Research On The Diagnosis Of COVID-19 Using CT Images Based On Self-adaptive Auxiliary Loss

The spread of COVID-19 has posed a huge threat to the lives and property of people around the world.Chest CT is considered as an effective tool for diagnosis and follow-up of COVID-19.However,reading CT images manually is time-consuming and laborious,and highly dependent on the radiologist’s clinical experience.For faster examination,au-tomatic COVID-19 diagnostic techniques using deep learning on CT images have received increasing attention.However,the number and category of existing datasets for COVID-19diagnosis that is available for using are limited.And the COVID-19 cases is less than the normal control’s,which leads to the problem of class imbalance.It makes the classification algorithms difficult to learn the discriminative boundaries since the data of some classes are rich while others are scarce.Therefore,training robust deep neural networks with imbal-anced data is a fundamental challenging but important task in the diagnosis of COVID-19.In this paper,advanced deep learning architecture is used to conduct in-depth research on automatic diagnosis of COVID-19.The main responsibibilities of our method include:(1)In this paper,an adaptive auxiliary loss is designed to monitor the imbalanced data.This loss function considers the effect of data overlap between CT slices,which is realized by quantifying the effective sample number as the weight of various types of cross entropy,and introduces reverse cross entropy by fully considering the possible mislabeling in clinical datasets.The parameters in the adaptive loss are obtained through network training without any manual adjustment.(2)In this paper,a novel and efficient classification pipeline called DSN-SAAL is developed.Considering that different depth of the network can learn diverse expression of characteristics,the supervised role of loss functions should be various in different stages of the network,which is reflected in the extent of partiality for minority classes in solving the problem of data imbalance.Our approach considers these issues simultaneously and constructs an adaptive auxiliary loss in the deep supervised network to effectively combine the learning hierarchy of distinct features in each phase of the model,and promote the feature learning of minority classes in the network,so as to make better diagnosis of COVID-19 in CT images.(3)We create a new COVID-19 diagnosis dataset(named COVID19-Diag)consisting of 6982 CT slices from 225 clinical cases in three categories: COVID-19,normal,and bacterial pneumonia.We have conducted series of experiments on COVID19-Diag to verify the effectiveness of DSN-SAAL under different degrees of imbalance from the whole and each component.At the same time,we evaluate DSN-SAAL on three additional publicly available COVID-19 datasets.The results show that DSN-SAAL outperforms the state-of-the-art methods and can achieve significant performance on generalization ability.